Module compilation 

GPU-native knowledge compilation.

This module is the home of GPU-native compilation + verification utilities.

Production correctness requires the GPU CDCL equivalence verifier (see validation).

Re-exports

pub use gpu_cnf::encode_cnf_gpu;

pub use gpu_cnf::GpuCnfEncoding;

pub use gpu_cnf::GpuCnfVarTables;

pub use gpu_d4::GpuCompileConfig;

pub use gpu_pir::GpuPirGraph;

pub use gpu_pir::GpuPirRoots;

pub use gpu_pir::PIR_AND;

pub use gpu_pir::PIR_LIT;

pub use gpu_pir::PIR_NEG_LIT;

pub use gpu_pir::PIR_OR;

pub use gpu_pir_intern::GpuPirInterner;

pub use gpu_pir_intern::PirBatch;

pub use gpu_weights::GpuWeights;

pub use gpu_weights::apply_query_vars_device;

pub use gpu_weights::build_evidence_by_var_gpu;

pub use gpu_weights::build_weights_gpu;

pub use gpu_weights::map_nodes_to_vars_gpu;

pub use gpu_weights::restore_query_vars_device;

pub use validation::build_equivalence_queries_gpu;

pub use validation::validate_equivalence_gpu;

pub use validation::validate_equivalence_gpu_gated;

pub use validation::GpuEquivalenceConfig;

pub use validation::GpuEquivalenceQueries;

Modules

disk_cache

On-disk circuit artifact cache.

gpu_cache

GPU-resident circuit cache helpers.

gpu_cnf

GPU-native Tseitin CNF encoding for PIR graphs.

gpu_d4

GPU-native Decision-DNNF knowledge compilation.

gpu_pir

GPU-resident Provenance IR (PIR) representation.

gpu_pir_intern

GPU PIR interner (device-side hash-consing).

gpu_weights

GPU-native weight table builders for exact inference.

sparse_matrix

CSR sparse matrix representation for CNF formulas.

validation

GPU-native equivalence validation (φ ≡ C) using the GPU CDCL verifier.

Structs

CircuitCompileProfile

Per-stage compilation timing (populated only when XLOG_WARMUP_PROFILE=1).

DeviceRandomVarList

Device-resident random-variable list for GPU smoothing.

Functions

compile_gpu_d4_and_verify

Compile CNF on GPU, then verify equivalence with GPU CDCL.

compile_gpu_d4_and_verify_cached

Compile CNF on GPU, cache the circuit, then verify equivalence with GPU CDCL.